See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/288183551

Forecasting Container Throughput at the Doraleh Port in Djibouti through

Time Series Analysis

Conference Paper · June 2015

DOI: 10.1142/9789814733878_0049

CITATION READS

1 839

2 authors:

Hawa Mohamed Ismael George Kobina VanDyck

Shanghai Maritime University Regional Maritime University
2 PUBLICATIONS   20 CITATIONS    8 PUBLICATIONS   42 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by George Kobina VanDyck on 26 December 2015.

The user has requested enhancement of the downloaded file.

 341

Forecasting Container Throughput at the Doraleh Port

Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

in Djibouti through Time Series Analysis

Hawa Mohamed Ismael

Logistics Research Center, Shanghai Maritime University,
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

Shanghai, 201306, China

E-mail: hawi_med_ismael@hotmail.com

George Kobina Vandyck

School of Economics and Management, Shanghai Maritime University,
Shanghai, 201306, China
E-mail: kobivandyck@yahoo.com

The Doraleh Container Terminal (DCT) located in Djibouti has been noted as the most
technologically advanced container terminal on the African continent. DCT’s strategic
location at the crossroads of the main shipping lanes connecting Asia, Africa and Europe
put it in a unique position to provide important shipping services to vessels plying that
route. This paper aims to forecast container throughput through the Doraleh Container
Port in Djibouti by Time Series Analysis. A selection of univariate forecasting models
has been used, namely Triple Exponential Smoothing Model, Grey Model and Linear
Regression Model. By utilizing the above three models and their combination, the
forecast of container throughput through the Doraleh port was realized. A comparison of
the different forecasting results of the three models, in addition to the combination
forecast is then undertaken, based on commonly used evaluation criteria Mean Absolute
Deviation (MAD) and Mean Absolute Percentage Error (MAPE). The study found that
the Linear Regression forecasting Model was the best prediction method for forecasting
the container throughput, since its forecast error was the least. Based on the regression
model, a ten (10) year forecast for container throughput at DCT has been made.
Keywords: Container Throughput Forecasting; Combination Forecasting Model;
Forecasting Accuracy.

1. Introduction
Opened in 2008, Doraleh Container Terminal (DCT) is the most technologically
advanced container terminal on the African continent. Just 11 km south of
the Autonomous Port of Djibouti, DCT enjoys the same strategic location at the
crossroads of the main shipping lanes connecting Asia, Africa and Europe. The
port lies on the major east-west trade route with minimal deviation and provides a
secure hub within the region for transshipment and relay activities [22].
 342

As in most ports around the world, containerization and globalization have

significantly impacted traffic volumes through the Doraleh Port in Djibouti. One
of the key issues for developing port facilities and construction of a new port is
information about the demand of container throughput. In port planning and
development, forecasting of container throughput demand is a necessary step in
predicting future revenues for proposed development projects. Hence, analysis
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

of container throughput demand is very important for port management.

Unfortunately, up to now, there is no published paper to the authors’ knowledge
that has dealt with forecasting the demand of container throughput in Djibouti.
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

2. Literature Review
Predicting container throughput at ports is very significant as such predictions
influence the development of a container port. The prediction techniques
available are both qualitative and quantitative methods. Qualitative methods for
forecasting include Delphi Method, the Subjective Probability Method and
Supply Survey Method. Alternatively, quantitative prediction methods include
Regression Analysis, Exponential Smoothing, Grey Model, the Neural Network
Method (NNM), Elastic Prediction Method and Combined Prediction Method.
The above models vary in the scope of application, complexity and accuracy.
Several factors have been noted to influence a port’s container throughput,
including the economic development of the port’s hinterland, the Gross
Domestic Product of the Economy and so on. There are several studies that have
been published on forecasting of port throughput by scholars. Below are some
relevant ones.
In previous literature [4] the combined model was applied composed of
grey-forecast model and Logistic-growth-curve model to improve the accuracy
of forecast model of cargo throughput for the port. The authors also use the
existing data of a current port to verify the validity of the combined model and
the results show that the combined model can obtain relatively higher forecast
accuracy more than any of the individual ones. Additionally, in a study on
container volumes through the port of Taiwan [15], it was identified that container
volumes were best forecast using classical time-series decomposition and that
more complex model did not necessarily produce more accurate results. In
another study, container throughput was analyzed by using cubic exponential
smoothing, GM (1, 1), and a combined forecast method and it was proved that
the combined model had the best precision [11]. Accordingly, different models
have been developed to design the best combination forecasts. Although, some
authors have observed that the combination forecasting does not necessarily lead
 343

to a better forecasting performance [1, 10], the averaging approach should be

used and interpreted with caution whereas “past model performance does not
always ensure future model performance” [2]. The linear regression model is
used in port throughput studies to identify and measures causal relationship
between variables determinants of throughput and forecasting, [6, 19] used the
linear regression method for throughput forecasting of ports in Hong Kong and
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

Taiwan.

3. Methodology
Regression Forecasting Model. Regression analysis is a collective name for
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

techniques for the modeling and analysis of numerical data consisting of values of
a dependent variable (also called response variable or measurement) and of one or
more independent variables (also known as explanatory variables or predictors)
[17]. The dependent variable in the regression equation is modeled as a function
of the independent variables, corresponding parameters (“constants”), and an
error part. The error part is treated as a random variable. It represents unexplained
variation in the dependent variable [20]. The parameters are estimated so as to
give the “best fit” of the data. Most commonly the best fit is evaluated by using
the least squares method, but other criteria have also been used.
Grey Forecast Model. Grey System theory [8, 18] offers a new approach to
deal mainly with the problems of uncertainty with few data points and/or poor
information which is said to be “partial known, partial unknown”. The basic
model of Grey model is GM (1, 1), a first-order differential model with one
input variable which has been successfully applied in many different researches.
It is obtained based on the following formula [5].
( )
= ( ( ) (1), ( ) (2), ( ) (3) … , ( ) ( ) , where n is the number of years
observed. The AGO formation of X (0) is defined as:
( )
= ( ( ) (1), ( ) (2), ( ) (3). . , ( ) ( )
(1)
^  ^ ^ ^ ^ ^
x ( k + 1) =  x (1) (1) − u/ a  e − a k + u/ a (1)
 
‘u’ and ‘a’ are the undetermined coefficient sequences.
The Grey theory doesn’t establish original data model, but generate data
model, so the Grey theory forecast data does not come directly from the
generated, but from the restored ones. We can restore the data in formula (2).
( ) ( ) ( )
()= ( )− ( − 1) ( = 2,3, . . ) (2)
 344

Triple Exponential Smoothing Forecasting Model. Exponential

Smoothing is a method for forecasting trends. The technique identifies historical
patterns of trend or seasonality in the data and then extrapolates these patterns
forward into the forecast period. Exponential Smoothing (or averaging”)
techniques are among the most widely used forecasting methods cargo
throughput [14].
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

Triple exponential smoothing method extends the double exponential

smoothing technique by including a smoothed multiplicative index to account
for the seasonal behavior of the forecast series.

S t(3) = αS t( 2) + (1 − α ) S t(−31) （ｔ＝１,２,３......n)

by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

(3)
at = 3S t(1) − 3S t( 2 ) + S t(3)
,

bt =
α
2(1 − α ) 2
[
(6 − 5α ) S t(1) − 2(5 − 4α ) S t( 2) + ( 4 − 3α ) S t(3) ]

ct =
α2
2(1 − α ) 2
[
S t(1) − 2 S t( 2 ) + St(3) ]
^
Yt + T = at + bt * T 2
+ c t *T (4)
Combination Forecasting Model. The idea of combination forecasts was
first described in a paper by [1]. The authors explained that a combination of
different single forecasting should be undertaken after considering the characters
of each single forecasting method. Combination forecasts can be generated
either by averaging different forecasts or using more sophisticated techniques,
including weighting or regression estimates [7, 9, 13, 16].
The accuracy and performance of the combination forecasting model has
been noted to be mainly associated with the performance consistency of each
individual forecasting model that has been utilized in addition to the assignment
of combination weights. Assuming “n” is the coefficient representing how
many ways to forecast the container throughput, y1 , y 2 …. y n , means the
predictive value, a 1 , a 2 …. a n , represent error for each single forecasting
method. The combination forecasting method is obtained as follows:
n
y= w y
i
i i (5)
 345

The combining weights w i are calculated to minimize the variance of the

error of the combined forecast, based on the assumption that each individual
forecast is unbiased. Mathematically, the weight is determined as follows:
n
(a1 + a 2 + .....a n − a i )
w i =1, w i =
(a1 + a 2 + .....a n ) * (n −`1)
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

4. Data and Results

Port container throughput data has been obtained from Djibouti Container
Terminal website to understand the growth pattern for both outbound and
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

inbound freight flow. The data is in thousand tons and is ranges from the year
2000 to 2013.
Economic Indicator Variable Data. The data for several economic
indicator variables are obtained from the World Bank data base. However, Gross
Domestic Product (GDP) is found suitable for the regression model and used to
estimate the future Container throughput. The major assumption is that
economic drivers are the key determinants of Container throughput (CT)
demand at seaports.
Table 1. Container throughput and GDP data.

Years CT [TEU] GDP [$]

2000 84104 551230862
2001 103098 572417441
2002 137978 591122040
2003 241151 622044666
2004 156603 666072102
2005 195250 708633195
2006 224896 768873684
2007 295746 847918929
2008 356462 999105339
2009 519500 1198997305
2010 406407 1128611700
2011 703617 1239144502
2012 743273 1353632942
2013 743794 1456344495
Source: Compilled from Port of Djibouti (2013) and World Bank (2012)

Triple Exponential Smoothing Forecast Model. As shown in the above

scatter plot, we can see the time series curve of container throughput shows a
very significant quadratic trend, so triple exponential smoothing is a reasonable
model for this study. Using the data from Table 1, the initial values are
established. The smoothing coefficient = 0.7 is estimated. The result is
shown Table 2 below.
 346

From the analysis, we obtained:

= 749448, = 98291, = −1470
^
The forecasting model formula therefore is Y13+T =749448+98291T-1470
T2.
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

Grey Model GM (1, 1). According to the Port historical data, we can
calculate through formula (1) for gray forecast model parameters: =
−0.160246574, = 97420.83079, while we can also get the small relative
error is between -0.25~0.31, and The distinguishing coefficient = 0.71 is
more than 0.5, so conclusively the prediction accuracy of this model is credible,
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

so that we can use this result as forecast model:

− 0.160246574 − 97420.83079

The Grey model GM (1, 1) for Djibouti Container Throughput Forecast is:
(1)
^  ^ ^ ^ ^ ^
x ( k + 1) =  x (1) (1) − u/ a  e − a k + u/ a =692047.2978 e 0.160246 k--607943.2978
 
Regression Model. In the model, container throughput is the dependent
variable, with GDP as the independent variable. For the linear regression
analysis, SPSS software was utilized. As can be seen from the above results, the
correlation coefficient between GDP and port throughput is 0.969, indicating
that the linear relationship between the economic indicators and the container
throughput is very close. R square statistic is 0.939, indicating the model
modestly fits the data. The container throughput and GDP is set respectively as,
y and x. The estimated regression equation can be expressed
= 0.001 − 311893.
Combined Forecast Model. Appropriate individual forecasting models
should be applied to develop the combination forecasting model. Since
combination forecasting model is usually constructed by 3~5 types of individual
models, this paper uses 3 types of individual methods; triple Exponential
Method, Grey (1, 1) Method and Regression method. The weight of Grey model,
exponential model, and regression model is set as , , . It can be calculated
from the formula (18):
= 0.22, = 0.38, = 0.40. Finally, combined forecast model is:
= 0.22 + 0.38 + 0.40
 347

5. Comparison of Forecasting Models

In previous studies, several measures of forecasting accuracy have been
proposed. Several studies and textbooks have recommended the use of the Mean
Absolute Percentage Error (MAPE) and Mean Absolute Deviation (MAD) in
ascertaining forecast accuracy. Therefore, the evaluation of the selected forecast
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

methods in order to improve forecast accuracy is an indispensable part this

forecasting study. This paper therefore compares the error of the selected
forecasting methods to verify the accuracy of the combined model.
Table 2. Actual and forecasting value of container throughput in Doraleh port.
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

Grey Triple Linear Combination

Years CT
Model Exponential Regression Model
2000 84104 84104 84104 89578 86294
2001 103098 120278 108393 109182 111323
2002 137978 141182 96880 127011 118679
2003 241151 165720 177130 150535 163982
2004 156603 194522 374381 179694 256973
2005 195250 228330 91681 218410 172435
2006 224896 268014 205035 242982 234069
2007 295746 314595 264251 280468 281814
2008 356462 369271 387692 382441 381539
2009 519500 433451 448616 560677 490104
2010 406407 508785 724729 532818 600457
2011 703617 597212 317546 604408 493817
2012 743273 701007 996275 681905 805568
2013 743793 822842 873014 751767 813478
MAPE % N/A 14% 34% 12% 18%
MAD N/A 46881 119418 39634 60246

We can see that the regression forecasting model is better than the others,
and it is closer to the true value. This indicates that the regression forecast model
during the port container throughput forecast is valid.
Table 3. Container throughput forecast based on regression model: 2014-2023.

Years CT [TEU] GDP [$]

2014 1,555,338,660.33 1,243,445.66
2015 1,655,539,605.71 1,343,646.61
2016 1,755,740,551.09 1,443,847.55
2017 1,855,941,496.47 1,544,048.50
2018 1,956,142,441.85 1,644,249.44
2019 2,056,343,387.22 1,744,450.39
2020 2,156,544,332.60 1,844,651.33
2021 2,256,745,277.98 1,944,852.28
2022 2,356,946,223.36 2,045,053.22
2023 2,457,147,168.78 2,145,254.17
 348

The regression forecast method forecast that the port container throughput
in 2018 will be 1.64 million tons. This will provide a basis for the Port to make
the port development strategy. In this case, the port throughput still has a close
relationship with the economic indicators.

6. Conclusion
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

The vision of the Port of Djibouti is to transform itself into an international hub
port. This will require significant strategic investment and planning. One
important consideration for such long-term investments is to be able to know
how demand will change in the years to come. This can only be achieved by a
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

reliable prediction model which will enable terminal operators to make timely
decisions on planning, construction and renovation of infrastructure and other
port facilities. As one of the key ports in East Africa for transit and
transshipment containerized cargo, the Port of Djibouti’s container throughput is
affected by various economic, social, political, and other factors. Foreign Direct
Investment, Population and GDP were chosen as the principle components to
analyze the port’s container throughput. It was established that GDP had a very
strong correlation to container throughput, and thus was used in the analysis.
The different models were compared and tested with observed container
throughput from 2000-2013. A comparison of the results showed that the
Regression model and the Grey model offered more accurate results than the
other two models. Moreover, the study concludes that better forecasting results
can be obtained from the regression model.

Acknowledgments
This work was supported in part by National Natural Science Foundation
of China (71171129) and Shanghai Science Commission Project (No.
12510501600, 14DZ2280200).

References
1. Bates, J.M. and Granger, C.W.J.: The Combination of Forecasts. Operational
Research Society, Vol. 20, No. 4, (1969), pp. 451-468.
2. Banternghansa, C. and McCracken, M.W.: Real-Time Forecast Averaging
with Alfred the Federal Reserve Bank of St. Louis, Working Paper
(2010)-033A, pp. 49-66.
3. Chen, S.H. and Chen, J.N.: Forecasting container throughputs at ports using
genetic programming, Expert Systems with Applications, Vol. 37, (2010),
pp. 2054-2058.
 349

4. Chi Zhang, Lei Huang, Zhichao Zhao: Research on combination forecast of

port cargo throughput based on time series and causality analysis, Journal of
Industrial Engineering and Management, Vol. 2, No. 1, (2013), pp. 124-134.
5. Chia-Nan Wang, Van-ThanhPhan: An Improvement the Accuracy of Grey
Forecasting Model for Cargo Throughput in International Commercial Ports
of Kaohsiung, Volume 11. Wseas Transactions on Business and Economics,
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

(2014).
6. Chou, C.C., Chu, C.W., and Liang, G.S.: A modified regression model for
forecasting the volumes of Taiwan’s import containers, Mathematical and
Computer Modeling, Vol. 47, (2008), pp. 797-807.
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

7. De Gooijer, J.G. and Hyndman, R.J.: 25 years of time series

forecasting. International Journal of Forecasting, Vol. 22, No. 3, (2006),
pp.443-473.
8. Deng, J.L.: Grey system theory tutorial. Huazhong University of Science and
Technology, Wuhan, China, (1990).
9. Goodwin, P.: New Evidence on the Value of Combining Forecasts.
Foresight: The International Journal of Applied Forecasting, No. 12, (2009),
pp. 33-35.
10. Kapetanios, G., Labhard, V. and Price, S.: Forecast combination and the
Bank of England’s suite of statistical forecasting models. Economic
Modeling, Vol. 25, No. 4, (2008), pp. 772-792.
11. Ma Xi-Chao, Wang Xuan, and Li Hai-feng: Throughput Forecast of Port
Container Based on Combined Model, Transport Standardization. 1 half of
No. 02/03. (2009), pp. 74-77.
12. Mahmoud, E.: Accuracy in Forecasting: A survey. Journal of Forecasting,
Vol. 3, No. 2, (1984), pp. 139-159.
13. Makridakis, S.: Why combining works? International Journal of Forecasting,
Vol. 5, No. 4, (1989), pp. 601-603.
14. Mark Hirschey: Managerial Economics, 12th edition, (2008).
15. Peng, W.Y. and Chu, C.W.: A comparison of univariate methods for
forcasting container throughput volumes, Mathematical and Computer
Moeling, Vol. 50, (2009), pp. 1045-1057.
16. Pesaran, M.H. and Pick, A.: Forecast Combination across Estimation
Widows. Journal of Business and Economic Statistics, Vol. 29, No. 2,
(2011), pp. 307-318.
17. Rodriguez, C.C.: Definition of regression (1996).
18. Sun, Y.M. and Zheng, G.P.: Research on Forecast of Port Cargo throughput
Capacity Based on Grey Theory, China Water Transport 4, (2007).
 350

19. Syafi’i: Multivariate Autoregressive Model for Forecasting the Demand of

Container Throughput in Indonesia, Media TeknikSipil, July (2006).
20. Weisberg, S.: Applied Linear Regression, John Wiley & Sons (2005).
21. William Seabrooke, Eddie C.M. Hui, William H.K. Lam, and Gordon K.C.
Wong: Forecasting cargo growth and regional role of the port of Hong Kong,
Department of Building and Real Estate, The Hong Kong Polytechnic
Applied Mechanics, Mechatronics and Intelligent Systems Downloaded from www.worldscientific.com

University, Hung Hom, Kowloon, Hong Kong, People’s Republic of China,

Vol. 20, No. 1, (2003), pp. 51-64.
22. World Bank Document: Study on regulation of private operators in the port
of Djibouti, June 2012.
by Mrs Hawa Mohamed Ismael on 12/19/15. For personal use only.

View publication stats